From a PTA Curve to a Decision
A probability of target attainment (PTA) can make dose options comparable, but a number alone is not a recommendation. “PTA is 92%” leaves open what target was assessed, in whom, under which regimen and assumptions, and whether 92% meets a prespecified decision rule. A clear communication keeps those elements together and distinguishes the simulation result from the choices used to interpret it.
For a dose regimen \(d\), the PTA is the probability, in a specified target population and under a specified model, that an exposure or response measure satisfies a defined target. In simulation, it is estimated by the proportion of simulated individuals who attain that target. The result is conditional on the model, its parameter values or uncertainty distribution, the simulated population, and the way the target is defined. It is not, by itself, an observed clinical response rate or a guarantee for any particular individual.
A report should make the following information readily visible:
- Target: the endpoint or exposure measure, threshold, and any relevant time window or other conditions.
- Population: who the simulation represents, including important covariate distributions and subgroup definitions.
- Decision rule: the PTA criterion and whether it applies to the overall population, each subgroup, or both.
- Uncertainty: Monte Carlo error, parameter uncertainty, and any other uncertainty evaluated, without treating them as interchangeable.
- Implementation: whether the proposed dose, interval, formulation, and administration are feasible.
State the Decision Rule Before Interpreting the Estimate
A PTA threshold is a decision criterion, not a hypothesis test. If a project specifies that the estimated PTA should be at least 90%, the criterion is \(\widehat{\mathrm{PTA}}(d)\geq 0.90\). That rule does not establish that a regimen is clinically effective in 90% of patients. It says that, under the stated model, target attainment is estimated for that proportion of the specified population. The clinical justification for the target and threshold must come from the relevant evidence and decision context.
The estimate also has simulation error. If simulated attainment indicators are independent Bernoulli outcomes with a common probability and model inputs are held fixed, the Monte Carlo standard error is estimated by
This standard error measures variation from using a finite number of simulated individuals. It does not measure uncertainty in the model parameters, uncertainty about the target threshold, or model misspecification. Those sources require separate evaluation; if parameter uncertainty is included in the simulation, the report should say how it was represented.
Worked Example: Reporting the PTA Criterion and Monte Carlo Error
Suppose a hypothetical simulation evaluates adults receiving one of three regimens. The target is a 24-hour exposure-to-susceptibility ratio of at least 100, and the prespecified decision criterion is an estimated PTA of at least 90%. For each regimen, 1,000 independent virtual individuals are simulated using one fixed set of model parameter estimates.
| Regimen | Attainers out of 1,000 | Estimated PTA |
|---|---|---|
| 500 mg every 12 hours | 760 | 76.0% |
| 750 mg every 12 hours | 880 | 88.0% |
| 1,000 mg every 24 hours | 934 | 93.4% |
For 1,000 mg every 24 hours, \(\widehat{\mathrm{PTA}}=934/1000=0.934\). Its estimated Monte Carlo standard error is
A rough 95% Monte Carlo interval using the normal approximation is \(0.934\pm1.96(0.00785)\), or approximately 0.919 to 0.949. This interval reflects finite-simulation error only, conditional on the fixed model inputs and the independent simulation assumption. The 1,000 mg regimen meets the point-estimate criterion; the other two evaluated regimens do not. Under a stricter rule requiring the lower end of this approximate Monte Carlo interval to exceed 90%, the 1,000 mg regimen also meets that rule. The report must identify which decision rule was chosen rather than silently switching between them.
These results support calling 1,000 mg every 24 hours the lowest-PTA-criterion-meeting regimen among those evaluated. They do not establish that it is the lowest possible dose meeting the criterion, since untested doses were not assessed. Nor does the interval account for uncertainty in the estimated model parameters.
Population Averages and Subgroup Performance
An overall PTA averages over the population composition used in the simulation. When the population contains distinct subgroups, a high overall result can conceal lower attainment in a subgroup. If subgroups are indexed by \(g\), \(w_g\) is the proportion of the target population in subgroup \(g\), and \(\mathrm{PTA}_g(d)\) is its PTA, then the overall population PTA is
This follows by partitioning the target-attainment event across mutually exclusive, exhaustive subgroups: \(P(A(d))=\sum_g P(A(d)\mid g)P(g)\). The subgroup weights must represent the population to which the recommendation is intended to apply. Equal weighting is appropriate only when the target population actually has equal subgroup proportions or when equal weighting is the explicit decision objective.
Worked Example: A Strong Overall PTA Can Hide a Weak Subgroup
In a hypothetical population, 60% have typical renal function and 40% have reduced renal function. At a tested regimen, 576 of 600 simulated individuals in the first subgroup and 336 of 400 in the second attain the target. The overall criterion is at least 90%, and the decision makers have also specified a minimum subgroup PTA of 85%.
The subgroup estimates are \(576/600=0.960\), or 96.0%, and \(336/400=0.840\), or 84.0%. Weighting by the stated population proportions gives
Thus the estimated overall PTA is 91.2%, above the 90% overall criterion, but the reduced-renal-function subgroup estimate is 84.0%, below its 85% criterion. The regimen does not meet both specified requirements. The subgroup Monte Carlo standard errors are approximately \(\sqrt{0.960(0.040)/600}=0.00800\) and \(\sqrt{0.840(0.160)/400}=0.0183\), respectively. The second subgroup estimate is less precise because its simulated sample is smaller and its attainment probability is lower.
A suitable communication would report both findings and state that the overall criterion is met while the subgroup criterion is not. Reporting only 91.2% would obscure information that changes the decision. Any next action—such as evaluating a different regimen or reconsidering the subgroup rule—requires an explicit rationale and additional analysis.
Separate Evidence, Uncertainty, and Assumptions
A decision statement should distinguish at least three layers. First is the simulation evidence: the estimated PTA and the population represented. Second is uncertainty: what changes when finite simulation size, model parameter estimates, or other specified assumptions vary. Third is the decision framework: the target, threshold, subgroup safeguards, and practical limits. These are related, but they are not the same kind of statement.
For example, a parameter-uncertainty analysis may produce a distribution of PTA estimates across parameter draws or refitted models. That distribution should be described by its method and summary, such as a median and stated quantiles. It should not be labelled a confidence interval unless the procedure and its interpretation justify that label. Likewise, a Monte Carlo interval around one PTA estimate should not be presented as if it captured uncertainty in the pharmacometric model.
Worked Example: Writing a Qualified Dose Recommendation
Consider a second hypothetical analysis of a population with a prespecified exposure target and a 90% PTA decision criterion. Among the tested regimens, 750 mg every 24 hours has an estimated PTA of 87%, and 1,000 mg every 24 hours has an estimated PTA of 91% in the primary simulation. The analysis also varies parameter estimates using a stated resampling procedure; across those analyses, the 10th and 90th percentiles of PTA for 1,000 mg are 86% and 95%. The intended setting has 250 mg tablets, a maximum daily dose of 1,000 mg, and no tablet splitting.
A transparent summary is: “In the simulated target population, 1,000 mg every 24 hours was the lowest tested regimen meeting the 90% PTA criterion in the primary analysis (estimated PTA 91%); 750 mg every 24 hours had an estimated PTA of 87%. Across the specified parameter-resampling analyses, the 10th–90th percentile range for the 1,000 mg regimen’s PTA was 86%–95%, indicating sensitivity to parameter uncertainty. The regimen can be administered as four 250 mg tablets once daily and is at, but does not exceed, the stated 1,000 mg daily limit. The result is conditional on the target definition, population, model, and resampling procedure.”
The wording identifies the selected regimen, the comparator result, the criterion, the uncertainty summary, and the formulation constraint. The 86%–95% range is not silently treated as a formal confidence interval, and the recommendation is not described as guaranteed to attain the target in 91% of treated individuals. If subgroup results or a toxicity limit are also decision-relevant, they must be reported before presenting the recommendation as acceptable.
Diagnostics, Alternatives, and Limits
Before communicating a recommendation, check that the reported PTA corresponds to the intended target and population, that the simulation denominator and subgroup weights are clear, and that the dose and interval match the regimen actually simulated. Compare subgroup and overall results, and identify whether the decision is based on a point estimate, a Monte Carlo bound, or an uncertainty analysis. Unexpectedly abrupt or nonmonotone PTA changes across doses, or results that change materially with simulation settings, warrant investigation rather than selective reporting.
PTA is useful when target attainment is the principal decision criterion and the target has a defensible interpretation. It is not a complete benefit-risk summary: a PTA threshold alone does not incorporate adverse outcomes, treatment burden, or patient preferences unless those are explicitly represented in the decision framework. A utility-based comparison can combine multiple valued outcomes, while a robustness analysis can show whether a recommendation persists across assumptions or subgroups. These alternatives answer broader or different questions; they do not make the target, population, or PTA estimate unnecessary to report.
- Mistake: “The 91% PTA means 91% of patients will respond.” Correction: It is a model-based estimate of target attainment in the stated population, not necessarily a clinical response probability.
- Mistake: Reporting only the overall PTA when a subgroup rule is part of the decision. Correction: Report subgroup estimates and state whether every specified criterion is met.
- Mistake: Calling Monte Carlo error the total uncertainty. Correction: State which uncertainty sources were evaluated and which remain outside the analysis.
- Mistake: Recommending a simulated regimen without checking available formulations or dose limits. Correction: Link the modeled regimen to a feasible administration plan, or state that feasibility has not been established.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A regimen attains the target in 450 of 500 simulations. What is its estimated PTA, and does it meet a prespecified 90% point-estimate criterion?
- Two subgroups make up 25% and 75% of a target population. Their PTAs are 80% and 96%. Calculate the population PTA.
- In one sentence, distinguish a Monte Carlo interval from uncertainty due to estimated model parameters.
- Why is it incorrect to interpret a PTA of 90% automatically as a 90% clinical response rate?
Solutions. (1) \(\widehat{\mathrm{PTA}}=450/500=0.900\), so it meets a criterion of at least 90%, exactly at the threshold. (2) \(0.25(0.80)+0.75(0.96)=0.20+0.72=0.92\), so the weighted population PTA is 92%. (3) A Monte Carlo interval describes finite-simulation error conditional on the inputs, whereas parameter uncertainty concerns uncertainty in the estimated model parameters and must be assessed separately. (4) PTA is the probability of attaining the specified model-based target in the simulated population; it equals a clinical response probability only if target attainment has been established as equivalent to that response and the relevant assumptions support that interpretation.